Managed Kubernetes

Nscale Kubernetes Service (NKS) lets you provision isolated, single-tenant Kubernetes clusters without managing the underlying infrastructure. Each cluster gets its own control plane and API server, so you can use standard tools like kubectl and helm straight away.

Prerequisite: You need an existing project before creating a cluster.

Availability

Managed Kubernetes is available in the reserved cloud service environment.

Requirements

Permissions and sufficient quota to create Kubernetes clusters
kubectl CLI installed locally

Cluster lifecycle

State	Description
Provisioning	The cluster is being created and resources are being allocated
Provisioned	The cluster is ready — kubeconfig is available for download
Deprovisioning	The cluster deletion is in progress and resources are being released
Error	Something went wrong during provisioning or operation

Actions you can take after creation:

Add a workload pool — add more node pools with different hardware configurations
Scale a workload pool — increase or decrease the node count in an existing pool
Delete a pool — remove an individual workload pool from the cluster
Delete the cluster — permanently remove the cluster and all its workload pools

Deleting a cluster is permanent. All workload pools, nodes, and ephemeral data are removed. Ensure any data you need is persisted externally before deleting.

Step-by-step

Create a cluster

In the Console, open your project, then go to Services → Kubernetes and click Create Cluster
Fill in the cluster details:
- Cluster Name — enter a name for your cluster
- Project — select the project the cluster should belong to
- Region — choose the region where the cluster will be deployed

The console provisions the control plane with default sizing (3 replicas and an automatically selected flavor). Control plane sizing cannot be changed in the console — to customize the flavor or replica count, create the cluster via the API instead (see Configure the control plane via API below).

Configure a workload pool Workload pools define the compute resources allocated to the cluster. For each pool, set:
- Workload Pool Name — must use lowercase alphanumeric characters and dashes (e.g. workload-pool-1)
- Node Type — hardware configuration (e.g. g.4.standard.80s or 1 × NVIDIA B200)
- Node Count — number of nodes in the pool
Review the cluster summary and click Create Cluster
- The cluster status will move to Provisioning, then Provisioned after a few minutes

Access your cluster

Once the cluster is Provisioned, the Access your cluster section becomes available in the cluster overview
- Click Download Kubeconfig to download kubeconfig.yaml, or Copy to clipboard to paste it into a file
Set your KUBECONFIG environment variable and verify connectivity:

export KUBECONFIG=~/Downloads/kubeconfig.yaml
kubectl get nodes

The output will show the nodes in your workload pool along with the Kubernetes version running on them.

Create a namespace for your workloads (optional)

kubectl create namespace dev-team
kubectl get namespaces

Configure the control plane via API

To specify control plane sizing when creating a cluster through the API, include the controlPlane object in your request:

curl -X POST "https://compute.nscale.com/api/v2/clusters" \
  -H "Authorization: Bearer $NSCALE_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "metadata": { "name": "my-cluster" },
    "spec": {
      "regionId": "<region-id>",
      "projectId": "<project-id>",
      "controlPlane": {
        "flavorId": "<flavor-id>",
        "replicas": 3
      },
      "workloadPools": [...]
    }
  }'

Field	Type	Required	Default	Description
`flavorId`	string	Yes (if `controlPlane` is set)	Auto-selected	Machine flavor for control plane nodes
`replicas`	integer	No	3	Number of control plane replicas (must be odd)

The replica count must be an odd number to maintain etcd quorum. Even values are rejected with a validation error.

Quotas

To check your current cluster quota, go to the Resource Usage section on the Dashboard. The dashboard shows your cluster usage alongside GPU, server, network, and filesystem quotas.

Common issues / troubleshooting

Access controls aren’t shown on the cluster page

Symptom: You don’t see Access your cluster (download/copy kubeconfig options)
Likely cause: The cluster is not yet provisioned; the access section appears after provisioning completes
Fix: Wait until the cluster status is Provisioned, then refresh the cluster overview page

Workload pool name validation error

Symptom: You can’t proceed when entering the workload pool name
Likely cause: The workload pool name must use lowercase alphanumeric characters and dashes
Fix: Rename the pool to match the required format (e.g. workload-pool-1)

Projects

Organise your clusters by project

Instances

Create individual VMs alongside your clusters

Terraform Provider

Manage clusters as infrastructure as code

API Reference

Manage clusters programmatically via the API

​Availability

​Requirements

​Cluster lifecycle

​Step-by-step

​Create a cluster

​Access your cluster

​Configure the control plane via API

​Quotas

​Common issues / troubleshooting

​Related resources

Projects

Instances

Terraform Provider

API Reference

Availability

Requirements

Cluster lifecycle

Step-by-step

Create a cluster

Access your cluster

Configure the control plane via API

Quotas

Common issues / troubleshooting

Related resources